Evaluation and comparison of open source program solutions for automatic seed counting on digital images

Seed number quantification is an essential agronomic parameter conducted mostly manually or by mechanical counters, both with obvious limitations. Digital image analysis provides a reliable and robust alternative to accurately calculate many biological features. This study presents and evaluates the performance of four open-source image-analysis programs i.e. ImageJ, CellProfiler, P-TRAP and SmartGrain to count crop seeds from digital images captured by camera and scanner. It also evaluates ImageJ program for automated seed counting using macro containing RenyiEntropy threshold algorithm. Digital images of cereal crop seeds were acquired i.e. wheat, barley, maize, rye, oat, sorghum, triticale and rice. All images contained 200 seeds per image present in an area of approx. 1400 cm2. RenyiEntropy threshold increased the seed count accuracy of ImageJ from digital camera images. Generally, seed counts from digital camera images of all crops were accurate, but software-crop combination had significant (p < 0.05) difference from reference value. Among image analysis programs, ImageJ produced mostly higher seed count across all observed crops than other programs. Mean seed counts from scanned images of maize were observed only by CellProfiler and P-TRAP, with other programs inappropriate due to high inaccuracy. These results suggest CellProfiler as a reliable image analysis program for seed counting from digital images. Benchmark test was also performed to compare speed of analysis. The automated seed count produced by image analysis programs described here allows faster, reliable and reproducible analysis, compared to standard manual method. To our knowledge this is the first study on using CellProfiler program for crop seed counting from digital images.